•Several effects of situation urgency on brake response time, previously observed separately, were replicated in one single simulator study.•Mechanistic models of brake timing, based on visual looming, were fitted to the human response times.•A looming threshold model was not able to capture the distribution of brake response times.•A model accumulating evidence from both visual looming and brake light onset provided the best fit to the data.•All models fitted the data better when quantifying looming cues as relative optical expansion rather than as optical expansion directly. Previous studies have shown the effect of a lead vehicle’s speed, deceleration rate and headway distance on drivers’ brake response times. However, how drivers perceive this information and use it to determine when to apply braking is still not quite clear. To better understand the underlying mechanisms, a driving simulator experiment was performed where each participant experienced nine deceleration scenarios. Previously reported effects of the lead vehicle’s speed, deceleration rate and headway distance on brake response time were firstly verified in this paper, using a multilevel model. Then, as an alternative to measures of speed, deceleration rate and distance, two visual looming-based metrics (angular expansion rate θ˙ of the lead vehicle on the driver’s retina, and inverse tau τ-1, the ratio between θ˙ and the optical size θ), considered to be more in line with typical human psycho-perceptual responses, were adopted to quantify situation urgency. These metrics were used in two previously proposed mechanistic models predicting brake onset: either when looming surpasses a threshold, or when the accumulated evidence (looming and other cues) reaches a threshold. Results showed that the looming threshold model did not capture the distribution of brake response time. However, regardless of looming metric, the accumulator models fitted the distribution of brake response times better than the pure threshold models. Accumulator models, including brake lights, provided a better model fit than looming-only versions. For all versions of the mechanistic models, models using τ-1 as the measure of looming fitted better than those using θ˙, indicating that the visual cues drivers used during rear-end collision avoidance may be more close to τ-1.